Research And Realization Of Character Segmentation Algorithm In Vehicle License Plate Recognition System

The Intelligent Transportation System(ITS) is becoming a hot research field as rapid increase in people's living standards day by day. As an important component part of the ITS, automatic vehicle license recognition system could real-time identify the vehicle license plate number, manage the vehicles, and thus it will improve the degree of automation of the monitoring and management in ITS. Therefore, the research of the automatic vehicle license plate recognition system has a very important practical value.The vehicle license plate recognition mainly includes three modules: vehicle license plate location, character segmentation and character recognition. This paper mainly does researches about the character segmentation module, and realizes the character segmentation system finally.There are three difficulties in the research, which are the binarization, the slant correction and the character segmentation. On the basis of the results of the analysis of domestic and foreign common related algorithms, and based on the vehicle license plate standards in China, this paper mainly does the following research work.1. Under the influence of obstacles,the range of the background gray-scale usually changes too much to distinguish the background and the target in the processing of binarization. In order to solve the problem, this paper binarizes the plate images using an improved local thresholding method, and obtain satisfactory binary images.2. The vehicle license plate images which obtain in the plate locating module are slant and need slant correction processing. Among these images, there are a large number ones which do not contain the license plate border. However, common Hough transform relies on the straight line too much, therefore the Hough transform does not applicable to the slant correction here. In this case, this paper detects the slant angle of the images first by the horizontal and vertical method, and then rotate the images through bilinear transformation. After experimental verification, we get good corrected images.3. In the part of character segmentation, taking into the account of relevant standards and the actual condition of the vehicle license plates, this paper proposed a character segmentation method to segment a plate image which contains seven characters. The method not only correctly split clear license plate images through experiments, but also the original slant plate images, the blurred license plate images and the plate images which surrounded by decorations, and so on. 4. Do experiments about the entire character segmentation module simulation. The experimental samples are 200 vehicle license plate images getting from the license plate locating module. It could correctly split 194 license plate images, and the correct rate is 97%.